Add restarting state and pipeline restart API

[mwylde]

This PR adds a new state Restarting and a new api POST /pipeline/{id}/restart which together allowing restarting pipelines. This can be used to restart a running pipeline (sometimes useful if it gets into some bad state that the system isn't able to automatically detect). But it can also be used for failed jobs, which currently cannot be recovered (without editing the db).

Now, a failed pipeline can be restarted via the API like this:

$ curl -XPOST localhost:8000/api/v1/pipelines/pl_OZWnhLx5I3/restart -d "{}" \
  -H "Content-Type: application/json"

There is also UI support to restart failed pipelines:

---

Restarting is a bit awkward to model in a reconciliation-based system like the controller; in general reconciliation requires a change to drive an operation. But for restarting, we want to force a state transition without actually changing any configuration.

This PR accomplishes that by adding restart_nonce fields on the job_configs and job_statuses tables. This is a number that can be incrementing in order to change the config while leaving everything the same.